Reading time: about 6–7 minutes.

Imagine growing a tiny cluster of human brain cells in a dish. The cells begin to fire electrical impulses. But when do many firing cells become a functioning network?

That is the central question of Functional Neural Networks in Human Brain Organoids, a review by Gu and colleagues. Its most useful idea is to study brain organoids at the network level: how neurons connect, coordinate their activity, and change that coordination in disease. The review brings together organoid culture methods, calcium imaging, electrical recordings, and disease-model experiments.

1. Growing neurons is only the first step

A brain organoid begins with human pluripotent stem cells (hPSCs). With appropriate culture conditions, those cells differentiate and self-organize into three-dimensional neural tissue. The resulting tissue can contain neural progenitors, excitatory and inhibitory neurons, and supporting glial cells.

These cell types play different roles. Excitatory neurons help propagate activity; inhibitory neurons, often using GABA signaling, regulate that activity; and astrocytes support the environment in which neural communication takes place. As neurons extend axons and dendrites, they form synapses. Repeated interactions can eventually produce collective patterns such as spikes, bursts, and synchronized oscillations.

Figure 2 — Gu et al. (2024)

Figure 2. Organoid development, cell types, and the emergence of network activity. Gu et al. (2024).

The review distinguishes two approaches to organoid development:

  • Unguided development: cells organize largely through intrinsic programs. This can produce diverse brain-region identities within one organoid, useful for examining interactions among regions.
  • Guided development: researchers supply specific small molecules and growth factors to steer differentiation toward a cortical, midbrain, or other regional identity. This offers a more consistent cellular composition for studying a particular circuit.

Over months, activity patterns also evolve. The review describes cortical organoids developing stronger firing, more frequent bursts, and more synchronized events; in an eight-month example, their spontaneous electrical-event timing resembles certain features of preterm neonatal EEG (review Figure 2H). This makes network maturation something researchers can measure rather than judge only from cellular appearance.

2. How do we know the neurons are communicating?

Researchers mainly use two complementary readouts. One watches intracellular calcium changes; the other listens to extracellular electrical activity.

Calcium imaging: watch cells light up

During neural activation, intracellular calcium can change. With fluorescent indicators such as GCaMP or Fluo-4, a microscope tracks each selected neuron’s changing brightness. Researchers commonly express this as ΔF/F, a change in fluorescence relative to a reference level.

Imagine observing 50 neurons. If several fluorescence traces repeatedly rise together, the group has coordinated calcium events. A spatial map can then show which cells tend to activate together. The review describes two-photon imaging experiments tracking calcium events across as many as 1,820 cells in an organoid network.

MEA: listen to electrical spikes

A microelectrode array (MEA) records voltage changes near neurons. Its major advantage is precise electrical timing: researchers can distinguish individual spikes, bursts, and the order of activity across recording sites.

One cited high-density CMOS-MEA system contains 26,400 electrode sites, with up to 1,024 simultaneously recorded channels. Researchers place organoid slices on the array, survey active locations, identify units from extracellular waveforms using a spike-sorting method such as Kilosort2, and then examine relationships between their spike trains.

Figure 3 — Gu et al. (2024)

Figure 3. A: calcium activity; B–C: MEA activity and single-unit maps; D: functional connectivity. Gu et al. (2024).

One useful network statistic is the spike time tiling coefficient (STTC). Its purpose is intuitive: ask whether spikes from two units occur close together more often than expected from their firing patterns. Repeat the comparison across many pairs, draw each unit as a node, and connect pairs with strong temporal coordination. The result is a functional connectivity graph.

This yields a practical analysis pipeline:

Raw recording → detected events → single-unit activity / calcium traces → pairwise coordination → network graph.

Instead of asking only ‘How often does an organoid fire?’, researchers can ask ‘Which groups fire together, how does their coordination develop, and how does an intervention reshape the network?’

3. Disease can change the rhythm of an entire network

Network readouts become particularly interesting when organoids are made from patient-derived cells or engineered to carry disease-associated mutations. The review presents two contrasting examples.

Rett syndrome: organoid experiments involving MECP2 mutations reveal large, synchronized calcium transients in cortical–ganglionic eminence fusion organoids. By comparing where the mutant cells were placed within the fused tissues, researchers investigated how altered inhibitory circuitry changes excitation–inhibition balance. The striking observation is excessive collective synchrony (review Figure 4A–C).

Alzheimer’s disease: in a cited study involving PSEN2-associated organoid models, healthy-derived neurons displayed more coordinated calcium events, while Alzheimer’s-associated neurons showed less synchronized activity (review Figure 4D–E). Here, the informative network feature is reduced coordination.

Figure 4 — Gu et al. (2024)

Figure 4. A–C: MECP2-associated hypersynchrony; D–E: healthy and AD-associated calcium activity. Gu et al. (2024).

The important move is from observing a single neuron to measuring a population phenotype: patterns of timing and connectivity reveal how a disease-associated change plays out across a circuit.

4. Why this matters for organoid intelligence

Once a living neural network produces measurable, input-dependent dynamics, it can also become part of an information-processing system.

The ingredients are straightforward: encode information as stimulation → let the organoid network respond → record activity with MEA → decode a useful output. The review cites the 2023 Brainoware work, where electrical inputs to an organoid were processed through its dynamic activity and a trained readout supported tasks such as speech recognition and nonlinear time-series prediction.

For this kind of system, firing rate alone is a small part of the story. Burst organization, response timing, synchrony, and changing functional connections provide a richer description of the network being used as the computational medium. The same measurement language also makes it possible to compare network states before and after stimulation or training.

Takeaway

A brain organoid becomes scientifically interesting as a neural network when its cells begin to coordinate, not simply fire. The review shows how to grow such networks, observe their electrical and calcium activity, and read meaningful differences in their collective dynamics. That is the bridge from a living tissue model to a measurable biological information-processing system.

References / 参考资料